1/* 2 * WMA compatible decoder 3 * Copyright (c) 2002 The Libav Project 4 * 5 * This file is part of Libav. 6 * 7 * Libav is free software; you can redistribute it and/or 8 * modify it under the terms of the GNU Lesser General Public 9 * License as published by the Free Software Foundation; either 10 * version 2.1 of the License, or (at your option) any later version. 11 * 12 * Libav is distributed in the hope that it will be useful, 13 * but WITHOUT ANY WARRANTY; without even the implied warranty of 14 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the GNU 15 * Lesser General Public License for more details. 16 * 17 * You should have received a copy of the GNU Lesser General Public 18 * License along with Libav; if not, write to the Free Software 19 * Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA 02110-1301 USA 20 */ 21 22/** 23 * @file 24 * WMA compatible decoder. 25 * This decoder handles Microsoft Windows Media Audio data, versions 1 & 2. 26 * WMA v1 is identified by audio format 0x160 in Microsoft media files 27 * (ASF/AVI/WAV). WMA v2 is identified by audio format 0x161. 28 * 29 * To use this decoder, a calling application must supply the extra data 30 * bytes provided with the WMA data. These are the extra, codec-specific 31 * bytes at the end of a WAVEFORMATEX data structure. Transmit these bytes 32 * to the decoder using the extradata[_size] fields in AVCodecContext. There 33 * should be 4 extra bytes for v1 data and 6 extra bytes for v2 data. 34 */ 35 36#include "avcodec.h" 37#include "wma.h" 38 39#undef NDEBUG 40#include <assert.h> 41 42#define EXPVLCBITS 8 43#define EXPMAX ((19+EXPVLCBITS-1)/EXPVLCBITS) 44 45#define HGAINVLCBITS 9 46#define HGAINMAX ((13+HGAINVLCBITS-1)/HGAINVLCBITS) 47 48static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len); 49 50#ifdef TRACE 51static void dump_shorts(WMACodecContext *s, const char *name, const short *tab, int n) 52{ 53 int i; 54 55 tprintf(s->avctx, "%s[%d]:\n", name, n); 56 for(i=0;i<n;i++) { 57 if ((i & 7) == 0) 58 tprintf(s->avctx, "%4d: ", i); 59 tprintf(s->avctx, " %5d.0", tab[i]); 60 if ((i & 7) == 7) 61 tprintf(s->avctx, "\n"); 62 } 63} 64 65static void dump_floats(WMACodecContext *s, const char *name, int prec, const float *tab, int n) 66{ 67 int i; 68 69 tprintf(s->avctx, "%s[%d]:\n", name, n); 70 for(i=0;i<n;i++) { 71 if ((i & 7) == 0) 72 tprintf(s->avctx, "%4d: ", i); 73 tprintf(s->avctx, " %8.*f", prec, tab[i]); 74 if ((i & 7) == 7) 75 tprintf(s->avctx, "\n"); 76 } 77 if ((i & 7) != 0) 78 tprintf(s->avctx, "\n"); 79} 80#endif 81 82static int wma_decode_init(AVCodecContext * avctx) 83{ 84 WMACodecContext *s = avctx->priv_data; 85 int i, flags2; 86 uint8_t *extradata; 87 88 if (!avctx->block_align) { 89 av_log(avctx, AV_LOG_ERROR, "block_align is not set\n"); 90 return AVERROR(EINVAL); 91 } 92 93 s->avctx = avctx; 94 95 /* extract flag infos */ 96 flags2 = 0; 97 extradata = avctx->extradata; 98 if (avctx->codec->id == CODEC_ID_WMAV1 && avctx->extradata_size >= 4) { 99 flags2 = AV_RL16(extradata+2); 100 } else if (avctx->codec->id == CODEC_ID_WMAV2 && avctx->extradata_size >= 6) { 101 flags2 = AV_RL16(extradata+4); 102 } 103// for(i=0; i<avctx->extradata_size; i++) 104// av_log(NULL, AV_LOG_ERROR, "%02X ", extradata[i]); 105 106 s->use_exp_vlc = flags2 & 0x0001; 107 s->use_bit_reservoir = flags2 & 0x0002; 108 s->use_variable_block_len = flags2 & 0x0004; 109 110 if(ff_wma_init(avctx, flags2)<0) 111 return -1; 112 113 /* init MDCT */ 114 for(i = 0; i < s->nb_block_sizes; i++) 115 ff_mdct_init(&s->mdct_ctx[i], s->frame_len_bits - i + 1, 1, 1.0); 116 117 if (s->use_noise_coding) { 118 init_vlc(&s->hgain_vlc, HGAINVLCBITS, sizeof(ff_wma_hgain_huffbits), 119 ff_wma_hgain_huffbits, 1, 1, 120 ff_wma_hgain_huffcodes, 2, 2, 0); 121 } 122 123 if (s->use_exp_vlc) { 124 init_vlc(&s->exp_vlc, EXPVLCBITS, sizeof(ff_aac_scalefactor_bits), //FIXME move out of context 125 ff_aac_scalefactor_bits, 1, 1, 126 ff_aac_scalefactor_code, 4, 4, 0); 127 } else { 128 wma_lsp_to_curve_init(s, s->frame_len); 129 } 130 131 avctx->sample_fmt = AV_SAMPLE_FMT_S16; 132 133 avcodec_get_frame_defaults(&s->frame); 134 avctx->coded_frame = &s->frame; 135 136 return 0; 137} 138 139/** 140 * compute x^-0.25 with an exponent and mantissa table. We use linear 141 * interpolation to reduce the mantissa table size at a small speed 142 * expense (linear interpolation approximately doubles the number of 143 * bits of precision). 144 */ 145static inline float pow_m1_4(WMACodecContext *s, float x) 146{ 147 union { 148 float f; 149 unsigned int v; 150 } u, t; 151 unsigned int e, m; 152 float a, b; 153 154 u.f = x; 155 e = u.v >> 23; 156 m = (u.v >> (23 - LSP_POW_BITS)) & ((1 << LSP_POW_BITS) - 1); 157 /* build interpolation scale: 1 <= t < 2. */ 158 t.v = ((u.v << LSP_POW_BITS) & ((1 << 23) - 1)) | (127 << 23); 159 a = s->lsp_pow_m_table1[m]; 160 b = s->lsp_pow_m_table2[m]; 161 return s->lsp_pow_e_table[e] * (a + b * t.f); 162} 163 164static void wma_lsp_to_curve_init(WMACodecContext *s, int frame_len) 165{ 166 float wdel, a, b; 167 int i, e, m; 168 169 wdel = M_PI / frame_len; 170 for(i=0;i<frame_len;i++) 171 s->lsp_cos_table[i] = 2.0f * cos(wdel * i); 172 173 /* tables for x^-0.25 computation */ 174 for(i=0;i<256;i++) { 175 e = i - 126; 176 s->lsp_pow_e_table[i] = pow(2.0, e * -0.25); 177 } 178 179 /* NOTE: these two tables are needed to avoid two operations in 180 pow_m1_4 */ 181 b = 1.0; 182 for(i=(1 << LSP_POW_BITS) - 1;i>=0;i--) { 183 m = (1 << LSP_POW_BITS) + i; 184 a = (float)m * (0.5 / (1 << LSP_POW_BITS)); 185 a = pow(a, -0.25); 186 s->lsp_pow_m_table1[i] = 2 * a - b; 187 s->lsp_pow_m_table2[i] = b - a; 188 b = a; 189 } 190} 191 192/** 193 * NOTE: We use the same code as Vorbis here 194 * @todo optimize it further with SSE/3Dnow 195 */ 196static void wma_lsp_to_curve(WMACodecContext *s, 197 float *out, float *val_max_ptr, 198 int n, float *lsp) 199{ 200 int i, j; 201 float p, q, w, v, val_max; 202 203 val_max = 0; 204 for(i=0;i<n;i++) { 205 p = 0.5f; 206 q = 0.5f; 207 w = s->lsp_cos_table[i]; 208 for(j=1;j<NB_LSP_COEFS;j+=2){ 209 q *= w - lsp[j - 1]; 210 p *= w - lsp[j]; 211 } 212 p *= p * (2.0f - w); 213 q *= q * (2.0f + w); 214 v = p + q; 215 v = pow_m1_4(s, v); 216 if (v > val_max) 217 val_max = v; 218 out[i] = v; 219 } 220 *val_max_ptr = val_max; 221} 222 223/** 224 * decode exponents coded with LSP coefficients (same idea as Vorbis) 225 */ 226static void decode_exp_lsp(WMACodecContext *s, int ch) 227{ 228 float lsp_coefs[NB_LSP_COEFS]; 229 int val, i; 230 231 for(i = 0; i < NB_LSP_COEFS; i++) { 232 if (i == 0 || i >= 8) 233 val = get_bits(&s->gb, 3); 234 else 235 val = get_bits(&s->gb, 4); 236 lsp_coefs[i] = ff_wma_lsp_codebook[i][val]; 237 } 238 239 wma_lsp_to_curve(s, s->exponents[ch], &s->max_exponent[ch], 240 s->block_len, lsp_coefs); 241} 242 243/** pow(10, i / 16.0) for i in -60..95 */ 244static const float pow_tab[] = { 245 1.7782794100389e-04, 2.0535250264571e-04, 246 2.3713737056617e-04, 2.7384196342644e-04, 247 3.1622776601684e-04, 3.6517412725484e-04, 248 4.2169650342858e-04, 4.8696752516586e-04, 249 5.6234132519035e-04, 6.4938163157621e-04, 250 7.4989420933246e-04, 8.6596432336006e-04, 251 1.0000000000000e-03, 1.1547819846895e-03, 252 1.3335214321633e-03, 1.5399265260595e-03, 253 1.7782794100389e-03, 2.0535250264571e-03, 254 2.3713737056617e-03, 2.7384196342644e-03, 255 3.1622776601684e-03, 3.6517412725484e-03, 256 4.2169650342858e-03, 4.8696752516586e-03, 257 5.6234132519035e-03, 6.4938163157621e-03, 258 7.4989420933246e-03, 8.6596432336006e-03, 259 1.0000000000000e-02, 1.1547819846895e-02, 260 1.3335214321633e-02, 1.5399265260595e-02, 261 1.7782794100389e-02, 2.0535250264571e-02, 262 2.3713737056617e-02, 2.7384196342644e-02, 263 3.1622776601684e-02, 3.6517412725484e-02, 264 4.2169650342858e-02, 4.8696752516586e-02, 265 5.6234132519035e-02, 6.4938163157621e-02, 266 7.4989420933246e-02, 8.6596432336007e-02, 267 1.0000000000000e-01, 1.1547819846895e-01, 268 1.3335214321633e-01, 1.5399265260595e-01, 269 1.7782794100389e-01, 2.0535250264571e-01, 270 2.3713737056617e-01, 2.7384196342644e-01, 271 3.1622776601684e-01, 3.6517412725484e-01, 272 4.2169650342858e-01, 4.8696752516586e-01, 273 5.6234132519035e-01, 6.4938163157621e-01, 274 7.4989420933246e-01, 8.6596432336007e-01, 275 1.0000000000000e+00, 1.1547819846895e+00, 276 1.3335214321633e+00, 1.5399265260595e+00, 277 1.7782794100389e+00, 2.0535250264571e+00, 278 2.3713737056617e+00, 2.7384196342644e+00, 279 3.1622776601684e+00, 3.6517412725484e+00, 280 4.2169650342858e+00, 4.8696752516586e+00, 281 5.6234132519035e+00, 6.4938163157621e+00, 282 7.4989420933246e+00, 8.6596432336007e+00, 283 1.0000000000000e+01, 1.1547819846895e+01, 284 1.3335214321633e+01, 1.5399265260595e+01, 285 1.7782794100389e+01, 2.0535250264571e+01, 286 2.3713737056617e+01, 2.7384196342644e+01, 287 3.1622776601684e+01, 3.6517412725484e+01, 288 4.2169650342858e+01, 4.8696752516586e+01, 289 5.6234132519035e+01, 6.4938163157621e+01, 290 7.4989420933246e+01, 8.6596432336007e+01, 291 1.0000000000000e+02, 1.1547819846895e+02, 292 1.3335214321633e+02, 1.5399265260595e+02, 293 1.7782794100389e+02, 2.0535250264571e+02, 294 2.3713737056617e+02, 2.7384196342644e+02, 295 3.1622776601684e+02, 3.6517412725484e+02, 296 4.2169650342858e+02, 4.8696752516586e+02, 297 5.6234132519035e+02, 6.4938163157621e+02, 298 7.4989420933246e+02, 8.6596432336007e+02, 299 1.0000000000000e+03, 1.1547819846895e+03, 300 1.3335214321633e+03, 1.5399265260595e+03, 301 1.7782794100389e+03, 2.0535250264571e+03, 302 2.3713737056617e+03, 2.7384196342644e+03, 303 3.1622776601684e+03, 3.6517412725484e+03, 304 4.2169650342858e+03, 4.8696752516586e+03, 305 5.6234132519035e+03, 6.4938163157621e+03, 306 7.4989420933246e+03, 8.6596432336007e+03, 307 1.0000000000000e+04, 1.1547819846895e+04, 308 1.3335214321633e+04, 1.5399265260595e+04, 309 1.7782794100389e+04, 2.0535250264571e+04, 310 2.3713737056617e+04, 2.7384196342644e+04, 311 3.1622776601684e+04, 3.6517412725484e+04, 312 4.2169650342858e+04, 4.8696752516586e+04, 313 5.6234132519035e+04, 6.4938163157621e+04, 314 7.4989420933246e+04, 8.6596432336007e+04, 315 1.0000000000000e+05, 1.1547819846895e+05, 316 1.3335214321633e+05, 1.5399265260595e+05, 317 1.7782794100389e+05, 2.0535250264571e+05, 318 2.3713737056617e+05, 2.7384196342644e+05, 319 3.1622776601684e+05, 3.6517412725484e+05, 320 4.2169650342858e+05, 4.8696752516586e+05, 321 5.6234132519035e+05, 6.4938163157621e+05, 322 7.4989420933246e+05, 8.6596432336007e+05, 323}; 324 325/** 326 * decode exponents coded with VLC codes 327 */ 328static int decode_exp_vlc(WMACodecContext *s, int ch) 329{ 330 int last_exp, n, code; 331 const uint16_t *ptr; 332 float v, max_scale; 333 uint32_t *q, *q_end, iv; 334 const float *ptab = pow_tab + 60; 335 const uint32_t *iptab = (const uint32_t*)ptab; 336 337 ptr = s->exponent_bands[s->frame_len_bits - s->block_len_bits]; 338 q = (uint32_t *)s->exponents[ch]; 339 q_end = q + s->block_len; 340 max_scale = 0; 341 if (s->version == 1) { 342 last_exp = get_bits(&s->gb, 5) + 10; 343 v = ptab[last_exp]; 344 iv = iptab[last_exp]; 345 max_scale = v; 346 n = *ptr++; 347 switch (n & 3) do { 348 case 0: *q++ = iv; 349 case 3: *q++ = iv; 350 case 2: *q++ = iv; 351 case 1: *q++ = iv; 352 } while ((n -= 4) > 0); 353 }else 354 last_exp = 36; 355 356 while (q < q_end) { 357 code = get_vlc2(&s->gb, s->exp_vlc.table, EXPVLCBITS, EXPMAX); 358 if (code < 0){ 359 av_log(s->avctx, AV_LOG_ERROR, "Exponent vlc invalid\n"); 360 return -1; 361 } 362 /* NOTE: this offset is the same as MPEG4 AAC ! */ 363 last_exp += code - 60; 364 if ((unsigned)last_exp + 60 >= FF_ARRAY_ELEMS(pow_tab)) { 365 av_log(s->avctx, AV_LOG_ERROR, "Exponent out of range: %d\n", 366 last_exp); 367 return -1; 368 } 369 v = ptab[last_exp]; 370 iv = iptab[last_exp]; 371 if (v > max_scale) 372 max_scale = v; 373 n = *ptr++; 374 switch (n & 3) do { 375 case 0: *q++ = iv; 376 case 3: *q++ = iv; 377 case 2: *q++ = iv; 378 case 1: *q++ = iv; 379 } while ((n -= 4) > 0); 380 } 381 s->max_exponent[ch] = max_scale; 382 return 0; 383} 384 385 386/** 387 * Apply MDCT window and add into output. 388 * 389 * We ensure that when the windows overlap their squared sum 390 * is always 1 (MDCT reconstruction rule). 391 */ 392static void wma_window(WMACodecContext *s, float *out) 393{ 394 float *in = s->output; 395 int block_len, bsize, n; 396 397 /* left part */ 398 if (s->block_len_bits <= s->prev_block_len_bits) { 399 block_len = s->block_len; 400 bsize = s->frame_len_bits - s->block_len_bits; 401 402 s->dsp.vector_fmul_add(out, in, s->windows[bsize], 403 out, block_len); 404 405 } else { 406 block_len = 1 << s->prev_block_len_bits; 407 n = (s->block_len - block_len) / 2; 408 bsize = s->frame_len_bits - s->prev_block_len_bits; 409 410 s->dsp.vector_fmul_add(out+n, in+n, s->windows[bsize], 411 out+n, block_len); 412 413 memcpy(out+n+block_len, in+n+block_len, n*sizeof(float)); 414 } 415 416 out += s->block_len; 417 in += s->block_len; 418 419 /* right part */ 420 if (s->block_len_bits <= s->next_block_len_bits) { 421 block_len = s->block_len; 422 bsize = s->frame_len_bits - s->block_len_bits; 423 424 s->dsp.vector_fmul_reverse(out, in, s->windows[bsize], block_len); 425 426 } else { 427 block_len = 1 << s->next_block_len_bits; 428 n = (s->block_len - block_len) / 2; 429 bsize = s->frame_len_bits - s->next_block_len_bits; 430 431 memcpy(out, in, n*sizeof(float)); 432 433 s->dsp.vector_fmul_reverse(out+n, in+n, s->windows[bsize], block_len); 434 435 memset(out+n+block_len, 0, n*sizeof(float)); 436 } 437} 438 439 440/** 441 * @return 0 if OK. 1 if last block of frame. return -1 if 442 * unrecorrable error. 443 */ 444static int wma_decode_block(WMACodecContext *s) 445{ 446 int n, v, a, ch, bsize; 447 int coef_nb_bits, total_gain; 448 int nb_coefs[MAX_CHANNELS]; 449 float mdct_norm; 450 FFTContext *mdct; 451 452#ifdef TRACE 453 tprintf(s->avctx, "***decode_block: %d:%d\n", s->frame_count - 1, s->block_num); 454#endif 455 456 /* compute current block length */ 457 if (s->use_variable_block_len) { 458 n = av_log2(s->nb_block_sizes - 1) + 1; 459 460 if (s->reset_block_lengths) { 461 s->reset_block_lengths = 0; 462 v = get_bits(&s->gb, n); 463 if (v >= s->nb_block_sizes){ 464 av_log(s->avctx, AV_LOG_ERROR, "prev_block_len_bits %d out of range\n", s->frame_len_bits - v); 465 return -1; 466 } 467 s->prev_block_len_bits = s->frame_len_bits - v; 468 v = get_bits(&s->gb, n); 469 if (v >= s->nb_block_sizes){ 470 av_log(s->avctx, AV_LOG_ERROR, "block_len_bits %d out of range\n", s->frame_len_bits - v); 471 return -1; 472 } 473 s->block_len_bits = s->frame_len_bits - v; 474 } else { 475 /* update block lengths */ 476 s->prev_block_len_bits = s->block_len_bits; 477 s->block_len_bits = s->next_block_len_bits; 478 } 479 v = get_bits(&s->gb, n); 480 if (v >= s->nb_block_sizes){ 481 av_log(s->avctx, AV_LOG_ERROR, "next_block_len_bits %d out of range\n", s->frame_len_bits - v); 482 return -1; 483 } 484 s->next_block_len_bits = s->frame_len_bits - v; 485 } else { 486 /* fixed block len */ 487 s->next_block_len_bits = s->frame_len_bits; 488 s->prev_block_len_bits = s->frame_len_bits; 489 s->block_len_bits = s->frame_len_bits; 490 } 491 492 /* now check if the block length is coherent with the frame length */ 493 s->block_len = 1 << s->block_len_bits; 494 if ((s->block_pos + s->block_len) > s->frame_len){ 495 av_log(s->avctx, AV_LOG_ERROR, "frame_len overflow\n"); 496 return -1; 497 } 498 499 if (s->nb_channels == 2) { 500 s->ms_stereo = get_bits1(&s->gb); 501 } 502 v = 0; 503 for(ch = 0; ch < s->nb_channels; ch++) { 504 a = get_bits1(&s->gb); 505 s->channel_coded[ch] = a; 506 v |= a; 507 } 508 509 bsize = s->frame_len_bits - s->block_len_bits; 510 511 /* if no channel coded, no need to go further */ 512 /* XXX: fix potential framing problems */ 513 if (!v) 514 goto next; 515 516 /* read total gain and extract corresponding number of bits for 517 coef escape coding */ 518 total_gain = 1; 519 for(;;) { 520 a = get_bits(&s->gb, 7); 521 total_gain += a; 522 if (a != 127) 523 break; 524 } 525 526 coef_nb_bits= ff_wma_total_gain_to_bits(total_gain); 527 528 /* compute number of coefficients */ 529 n = s->coefs_end[bsize] - s->coefs_start; 530 for(ch = 0; ch < s->nb_channels; ch++) 531 nb_coefs[ch] = n; 532 533 /* complex coding */ 534 if (s->use_noise_coding) { 535 536 for(ch = 0; ch < s->nb_channels; ch++) { 537 if (s->channel_coded[ch]) { 538 int i, n, a; 539 n = s->exponent_high_sizes[bsize]; 540 for(i=0;i<n;i++) { 541 a = get_bits1(&s->gb); 542 s->high_band_coded[ch][i] = a; 543 /* if noise coding, the coefficients are not transmitted */ 544 if (a) 545 nb_coefs[ch] -= s->exponent_high_bands[bsize][i]; 546 } 547 } 548 } 549 for(ch = 0; ch < s->nb_channels; ch++) { 550 if (s->channel_coded[ch]) { 551 int i, n, val, code; 552 553 n = s->exponent_high_sizes[bsize]; 554 val = (int)0x80000000; 555 for(i=0;i<n;i++) { 556 if (s->high_band_coded[ch][i]) { 557 if (val == (int)0x80000000) { 558 val = get_bits(&s->gb, 7) - 19; 559 } else { 560 code = get_vlc2(&s->gb, s->hgain_vlc.table, HGAINVLCBITS, HGAINMAX); 561 if (code < 0){ 562 av_log(s->avctx, AV_LOG_ERROR, "hgain vlc invalid\n"); 563 return -1; 564 } 565 val += code - 18; 566 } 567 s->high_band_values[ch][i] = val; 568 } 569 } 570 } 571 } 572 } 573 574 /* exponents can be reused in short blocks. */ 575 if ((s->block_len_bits == s->frame_len_bits) || 576 get_bits1(&s->gb)) { 577 for(ch = 0; ch < s->nb_channels; ch++) { 578 if (s->channel_coded[ch]) { 579 if (s->use_exp_vlc) { 580 if (decode_exp_vlc(s, ch) < 0) 581 return -1; 582 } else { 583 decode_exp_lsp(s, ch); 584 } 585 s->exponents_bsize[ch] = bsize; 586 } 587 } 588 } 589 590 /* parse spectral coefficients : just RLE encoding */ 591 for(ch = 0; ch < s->nb_channels; ch++) { 592 if (s->channel_coded[ch]) { 593 int tindex; 594 WMACoef* ptr = &s->coefs1[ch][0]; 595 596 /* special VLC tables are used for ms stereo because 597 there is potentially less energy there */ 598 tindex = (ch == 1 && s->ms_stereo); 599 memset(ptr, 0, s->block_len * sizeof(WMACoef)); 600 ff_wma_run_level_decode(s->avctx, &s->gb, &s->coef_vlc[tindex], 601 s->level_table[tindex], s->run_table[tindex], 602 0, ptr, 0, nb_coefs[ch], 603 s->block_len, s->frame_len_bits, coef_nb_bits); 604 } 605 if (s->version == 1 && s->nb_channels >= 2) { 606 align_get_bits(&s->gb); 607 } 608 } 609 610 /* normalize */ 611 { 612 int n4 = s->block_len / 2; 613 mdct_norm = 1.0 / (float)n4; 614 if (s->version == 1) { 615 mdct_norm *= sqrt(n4); 616 } 617 } 618 619 /* finally compute the MDCT coefficients */ 620 for(ch = 0; ch < s->nb_channels; ch++) { 621 if (s->channel_coded[ch]) { 622 WMACoef *coefs1; 623 float *coefs, *exponents, mult, mult1, noise; 624 int i, j, n, n1, last_high_band, esize; 625 float exp_power[HIGH_BAND_MAX_SIZE]; 626 627 coefs1 = s->coefs1[ch]; 628 exponents = s->exponents[ch]; 629 esize = s->exponents_bsize[ch]; 630 mult = pow(10, total_gain * 0.05) / s->max_exponent[ch]; 631 mult *= mdct_norm; 632 coefs = s->coefs[ch]; 633 if (s->use_noise_coding) { 634 mult1 = mult; 635 /* very low freqs : noise */ 636 for(i = 0;i < s->coefs_start; i++) { 637 *coefs++ = s->noise_table[s->noise_index] * 638 exponents[i<<bsize>>esize] * mult1; 639 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); 640 } 641 642 n1 = s->exponent_high_sizes[bsize]; 643 644 /* compute power of high bands */ 645 exponents = s->exponents[ch] + 646 (s->high_band_start[bsize]<<bsize>>esize); 647 last_high_band = 0; /* avoid warning */ 648 for(j=0;j<n1;j++) { 649 n = s->exponent_high_bands[s->frame_len_bits - 650 s->block_len_bits][j]; 651 if (s->high_band_coded[ch][j]) { 652 float e2, v; 653 e2 = 0; 654 for(i = 0;i < n; i++) { 655 v = exponents[i<<bsize>>esize]; 656 e2 += v * v; 657 } 658 exp_power[j] = e2 / n; 659 last_high_band = j; 660 tprintf(s->avctx, "%d: power=%f (%d)\n", j, exp_power[j], n); 661 } 662 exponents += n<<bsize>>esize; 663 } 664 665 /* main freqs and high freqs */ 666 exponents = s->exponents[ch] + (s->coefs_start<<bsize>>esize); 667 for(j=-1;j<n1;j++) { 668 if (j < 0) { 669 n = s->high_band_start[bsize] - 670 s->coefs_start; 671 } else { 672 n = s->exponent_high_bands[s->frame_len_bits - 673 s->block_len_bits][j]; 674 } 675 if (j >= 0 && s->high_band_coded[ch][j]) { 676 /* use noise with specified power */ 677 mult1 = sqrt(exp_power[j] / exp_power[last_high_band]); 678 /* XXX: use a table */ 679 mult1 = mult1 * pow(10, s->high_band_values[ch][j] * 0.05); 680 mult1 = mult1 / (s->max_exponent[ch] * s->noise_mult); 681 mult1 *= mdct_norm; 682 for(i = 0;i < n; i++) { 683 noise = s->noise_table[s->noise_index]; 684 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); 685 *coefs++ = noise * 686 exponents[i<<bsize>>esize] * mult1; 687 } 688 exponents += n<<bsize>>esize; 689 } else { 690 /* coded values + small noise */ 691 for(i = 0;i < n; i++) { 692 noise = s->noise_table[s->noise_index]; 693 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); 694 *coefs++ = ((*coefs1++) + noise) * 695 exponents[i<<bsize>>esize] * mult; 696 } 697 exponents += n<<bsize>>esize; 698 } 699 } 700 701 /* very high freqs : noise */ 702 n = s->block_len - s->coefs_end[bsize]; 703 mult1 = mult * exponents[((-1<<bsize))>>esize]; 704 for(i = 0; i < n; i++) { 705 *coefs++ = s->noise_table[s->noise_index] * mult1; 706 s->noise_index = (s->noise_index + 1) & (NOISE_TAB_SIZE - 1); 707 } 708 } else { 709 /* XXX: optimize more */ 710 for(i = 0;i < s->coefs_start; i++) 711 *coefs++ = 0.0; 712 n = nb_coefs[ch]; 713 for(i = 0;i < n; i++) { 714 *coefs++ = coefs1[i] * exponents[i<<bsize>>esize] * mult; 715 } 716 n = s->block_len - s->coefs_end[bsize]; 717 for(i = 0;i < n; i++) 718 *coefs++ = 0.0; 719 } 720 } 721 } 722 723#ifdef TRACE 724 for(ch = 0; ch < s->nb_channels; ch++) { 725 if (s->channel_coded[ch]) { 726 dump_floats(s, "exponents", 3, s->exponents[ch], s->block_len); 727 dump_floats(s, "coefs", 1, s->coefs[ch], s->block_len); 728 } 729 } 730#endif 731 732 if (s->ms_stereo && s->channel_coded[1]) { 733 /* nominal case for ms stereo: we do it before mdct */ 734 /* no need to optimize this case because it should almost 735 never happen */ 736 if (!s->channel_coded[0]) { 737 tprintf(s->avctx, "rare ms-stereo case happened\n"); 738 memset(s->coefs[0], 0, sizeof(float) * s->block_len); 739 s->channel_coded[0] = 1; 740 } 741 742 s->dsp.butterflies_float(s->coefs[0], s->coefs[1], s->block_len); 743 } 744 745next: 746 mdct = &s->mdct_ctx[bsize]; 747 748 for(ch = 0; ch < s->nb_channels; ch++) { 749 int n4, index; 750 751 n4 = s->block_len / 2; 752 if(s->channel_coded[ch]){ 753 mdct->imdct_calc(mdct, s->output, s->coefs[ch]); 754 }else if(!(s->ms_stereo && ch==1)) 755 memset(s->output, 0, sizeof(s->output)); 756 757 /* multiply by the window and add in the frame */ 758 index = (s->frame_len / 2) + s->block_pos - n4; 759 wma_window(s, &s->frame_out[ch][index]); 760 } 761 762 /* update block number */ 763 s->block_num++; 764 s->block_pos += s->block_len; 765 if (s->block_pos >= s->frame_len) 766 return 1; 767 else 768 return 0; 769} 770 771/* decode a frame of frame_len samples */ 772static int wma_decode_frame(WMACodecContext *s, int16_t *samples) 773{ 774 int ret, n, ch, incr; 775 const float *output[MAX_CHANNELS]; 776 777#ifdef TRACE 778 tprintf(s->avctx, "***decode_frame: %d size=%d\n", s->frame_count++, s->frame_len); 779#endif 780 781 /* read each block */ 782 s->block_num = 0; 783 s->block_pos = 0; 784 for(;;) { 785 ret = wma_decode_block(s); 786 if (ret < 0) 787 return -1; 788 if (ret) 789 break; 790 } 791 792 /* convert frame to integer */ 793 n = s->frame_len; 794 incr = s->nb_channels; 795 for (ch = 0; ch < MAX_CHANNELS; ch++) 796 output[ch] = s->frame_out[ch]; 797 s->fmt_conv.float_to_int16_interleave(samples, output, n, incr); 798 for (ch = 0; ch < incr; ch++) { 799 /* prepare for next block */ 800 memmove(&s->frame_out[ch][0], &s->frame_out[ch][n], n * sizeof(float)); 801 } 802 803#ifdef TRACE 804 dump_shorts(s, "samples", samples, n * s->nb_channels); 805#endif 806 return 0; 807} 808 809static int wma_decode_superframe(AVCodecContext *avctx, void *data, 810 int *got_frame_ptr, AVPacket *avpkt) 811{ 812 const uint8_t *buf = avpkt->data; 813 int buf_size = avpkt->size; 814 WMACodecContext *s = avctx->priv_data; 815 int nb_frames, bit_offset, i, pos, len, ret; 816 uint8_t *q; 817 int16_t *samples; 818 819 tprintf(avctx, "***decode_superframe:\n"); 820 821 if(buf_size==0){ 822 s->last_superframe_len = 0; 823 return 0; 824 } 825 if (buf_size < s->block_align) { 826 av_log(avctx, AV_LOG_ERROR, 827 "Input packet size too small (%d < %d)\n", 828 buf_size, s->block_align); 829 return AVERROR_INVALIDDATA; 830 } 831 buf_size = s->block_align; 832 833 init_get_bits(&s->gb, buf, buf_size*8); 834 835 if (s->use_bit_reservoir) { 836 /* read super frame header */ 837 skip_bits(&s->gb, 4); /* super frame index */ 838 nb_frames = get_bits(&s->gb, 4) - (s->last_superframe_len <= 0); 839 } else { 840 nb_frames = 1; 841 } 842 843 /* get output buffer */ 844 s->frame.nb_samples = nb_frames * s->frame_len; 845 if ((ret = avctx->get_buffer(avctx, &s->frame)) < 0) { 846 av_log(avctx, AV_LOG_ERROR, "get_buffer() failed\n"); 847 return ret; 848 } 849 samples = (int16_t *)s->frame.data[0]; 850 851 if (s->use_bit_reservoir) { 852 bit_offset = get_bits(&s->gb, s->byte_offset_bits + 3); 853 if (bit_offset > get_bits_left(&s->gb)) { 854 av_log(avctx, AV_LOG_ERROR, 855 "Invalid last frame bit offset %d > buf size %d (%d)\n", 856 bit_offset, get_bits_left(&s->gb), buf_size); 857 goto fail; 858 } 859 860 if (s->last_superframe_len > 0) { 861 // printf("skip=%d\n", s->last_bitoffset); 862 /* add bit_offset bits to last frame */ 863 if ((s->last_superframe_len + ((bit_offset + 7) >> 3)) > 864 MAX_CODED_SUPERFRAME_SIZE) 865 goto fail; 866 q = s->last_superframe + s->last_superframe_len; 867 len = bit_offset; 868 while (len > 7) { 869 *q++ = (get_bits)(&s->gb, 8); 870 len -= 8; 871 } 872 if (len > 0) { 873 *q++ = (get_bits)(&s->gb, len) << (8 - len); 874 } 875 memset(q, 0, FF_INPUT_BUFFER_PADDING_SIZE); 876 877 /* XXX: bit_offset bits into last frame */ 878 init_get_bits(&s->gb, s->last_superframe, s->last_superframe_len * 8 + bit_offset); 879 /* skip unused bits */ 880 if (s->last_bitoffset > 0) 881 skip_bits(&s->gb, s->last_bitoffset); 882 /* this frame is stored in the last superframe and in the 883 current one */ 884 if (wma_decode_frame(s, samples) < 0) 885 goto fail; 886 samples += s->nb_channels * s->frame_len; 887 nb_frames--; 888 } 889 890 /* read each frame starting from bit_offset */ 891 pos = bit_offset + 4 + 4 + s->byte_offset_bits + 3; 892 if (pos >= MAX_CODED_SUPERFRAME_SIZE * 8 || pos > buf_size * 8) 893 return AVERROR_INVALIDDATA; 894 init_get_bits(&s->gb, buf + (pos >> 3), (buf_size - (pos >> 3))*8); 895 len = pos & 7; 896 if (len > 0) 897 skip_bits(&s->gb, len); 898 899 s->reset_block_lengths = 1; 900 for(i=0;i<nb_frames;i++) { 901 if (wma_decode_frame(s, samples) < 0) 902 goto fail; 903 samples += s->nb_channels * s->frame_len; 904 } 905 906 /* we copy the end of the frame in the last frame buffer */ 907 pos = get_bits_count(&s->gb) + ((bit_offset + 4 + 4 + s->byte_offset_bits + 3) & ~7); 908 s->last_bitoffset = pos & 7; 909 pos >>= 3; 910 len = buf_size - pos; 911 if (len > MAX_CODED_SUPERFRAME_SIZE || len < 0) { 912 av_log(s->avctx, AV_LOG_ERROR, "len %d invalid\n", len); 913 goto fail; 914 } 915 s->last_superframe_len = len; 916 memcpy(s->last_superframe, buf + pos, len); 917 } else { 918 /* single frame decode */ 919 if (wma_decode_frame(s, samples) < 0) 920 goto fail; 921 samples += s->nb_channels * s->frame_len; 922 } 923 924//av_log(NULL, AV_LOG_ERROR, "%d %d %d %d outbytes:%d eaten:%d\n", s->frame_len_bits, s->block_len_bits, s->frame_len, s->block_len, (int8_t *)samples - (int8_t *)data, s->block_align); 925 926 *got_frame_ptr = 1; 927 *(AVFrame *)data = s->frame; 928 929 return s->block_align; 930 fail: 931 /* when error, we reset the bit reservoir */ 932 s->last_superframe_len = 0; 933 return -1; 934} 935 936static av_cold void flush(AVCodecContext *avctx) 937{ 938 WMACodecContext *s = avctx->priv_data; 939 940 s->last_bitoffset= 941 s->last_superframe_len= 0; 942} 943 944AVCodec ff_wmav1_decoder = { 945 .name = "wmav1", 946 .type = AVMEDIA_TYPE_AUDIO, 947 .id = CODEC_ID_WMAV1, 948 .priv_data_size = sizeof(WMACodecContext), 949 .init = wma_decode_init, 950 .close = ff_wma_end, 951 .decode = wma_decode_superframe, 952 .flush = flush, 953 .capabilities = CODEC_CAP_DR1, 954 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 1"), 955}; 956 957AVCodec ff_wmav2_decoder = { 958 .name = "wmav2", 959 .type = AVMEDIA_TYPE_AUDIO, 960 .id = CODEC_ID_WMAV2, 961 .priv_data_size = sizeof(WMACodecContext), 962 .init = wma_decode_init, 963 .close = ff_wma_end, 964 .decode = wma_decode_superframe, 965 .flush = flush, 966 .capabilities = CODEC_CAP_DR1, 967 .long_name = NULL_IF_CONFIG_SMALL("Windows Media Audio 2"), 968}; 969